---

Home

Basic Concepts White Paper

3 Minute Overview Video

BrandLoyalties Full Presentation

"Big Data Meets Deep Learning" Video

BrandLoyalties FAQs

---

Frequently Asked Questions:

What is BrandLoyalties?

What are the BrandLoyalties metrics?

Why is "Brand Loyalty" important?

Have "brand loyalties" ever been used successfully in the past to build investment portfolios?

How do your new metrics compare with what Peter Lynch was doing?

What makes your new BrandLoyalties metrics unique?

How can this benefit portfolio managers?

How do you guys do it?

How long have you been doing this? How far back does your data go?

What are the caveats to using your data?

If the "Signal Relevance" of the BrandLoyalties data changes over time, how can we quantify and monitor it?

What are the criteria for the rankings?

Is your data measuring sentiment?

From what web sites are your citations collected?

When is your data collected? And how soon is it provided to your clients?

Do you have data available for back-testing?

How often do you revise the list of brands that you are tracking?

What equities are in the BrandLoyalties "universe"?

In what data formats are the metrics available?

Do you have any suggestions on the best way to get started when using Artificial Intelligence and Machine Learning with your data?

How effective are the BrandLoyalties metrics?

What other information do you provide?

What is BrandLoyalties, Inc. and what is its history?

How are the BrandLoyalties subscription plans structured and priced?

---

What is BrandLoyalties?

BrandLoyalties, Inc. is a provider of alternative investment research based on the daily analysis of consumer behavior found on-line in "Big Data". We provide a daily suite of ticker-level metrics for over 2,500 US equities (and a total of over 4,500 equities globally) to hedge funds and institutional asset managers.


What are the BrandLoyalties metrics?

We start with daily measurements of the "Web Luminosity" of the brand names owned by publicly traded corporations. The loyalty of on-line consumers to those brands can then be determined by how a corporation's "Web Luminosity" shifts over time. Those shifts in the loyalties that consumers have for the brands of corporations are then translated into our "brand loyalties" metrics.


Why is "Brand Loyalty" important?

Brand loyalty was one of the first intangible assets recognized in academic literature. This asset is of key interest to investors because of the value that "brand loyalty" generates to companies in terms of:

-- a substantial entry barrier to competitors;

-- an increase in the firm's ability to respond to competitive threats;

-- greater sales and revenue; and

-- a customer base less sensitive to the marketing efforts of competitors.

Jack Treynor (2002) studied the concept in more blatantly economic terms, arguing that "Brand loyalty manifests itself in consumers' willingness to pay a higher price for the brand they prefer." He went on to say, "The importance of brand loyalty stems from the anxiety of consumers at a point when the industry is unstable and certain brand names hold a better reputation for the quality of their product than others in the industry." Treynor used historic rates of change in sales among competitors (i.e., changing market share) in attempting to find a quantitative proxy for brand loyalty.


Have "brand loyalties" ever been used successfully in the past to build investment portfolios?

Savvy investors have always been interested in gauging consumer loyalties. Traditionally, consumer surveys, focus groups and even retail traffic counters have been deployed to understand whether certain stores, brands and products may be gaining or losing market share. The most obvious historical example is the legendary performance of Peter Lynch and the Magellan Fund during 1977-1990, when the average annual return experienced by its shareholders was over 29%. One of Peter's key equity screens involved initially selecting those companies that were selling products enjoyed again and again by his family, neighbors and colleagues. He then did his subsequent rigorous fundamental analysis on the companies selected by what he called "the power of common knowledge" (i.e., brand loyalty).

At BrandLoyalties, Inc. we took a look at his results and asked several basic questions: What has changed since 1977? How can we harness evolving markets and twenty-first century technologies to improve the scope, timeliness and quantitative accuracy of Peter's methodologies? How can we adopt his approach and adapt it for the 21st century?


How do your new metrics compare with what Peter Lynch was doing?

We have simply expanded his concept by capturing "Big Data" from tens of millions of people shopping on the web, and doing the sampling in "real-time." As a consequence we have a twenty-first century version of his "brand loyalties" focus groups that has:

-- Unprecedented timeliness, with day-by-day results that are always current through yesterday.

-- High granularity, with results that are actionable at the equity-by-equity level.

-- Extensive scope and coverage, with over 2,500 global equities tracked (including over 1,800 US mid and large cap equities).

-- Daily updates to each equity's "brand loyalty," reflecting a mix of year-over-year and trailing 90 day shifts in customer affections.


What makes your new BrandLoyalties metrics unique?

Our "brand loyalties" metrics track actual on-line consumer behavior, and for that reason they are purely quantitative. We are using proprietary state-of-the-art technologies embedded in thousands of cloud based servers to extract consumer citations of brand names within "Big Data". Our data captures over 100,000,000 on-line consumer brand references daily. This methodology has vastly greater sample sizes than conventional questionnaire surveys or boutique focus groups. The resulting metrics are also orders of magnitude more current than other sources, with daily updates posted on the subscriber portion of our website every morning. This is a new and unique set of metrics that is unprecedented, unrivaled and exclusively available at www.BrandLoyalties.com.


How can this benefit portfolio managers?

Now portfolio managers can tell on a daily basis whether the equities currently in their portfolios have customers flocking to them or (conversely) avoiding them like the plague. Furthermore, they will know this well before earnings season -- and quite possibly even before the corporate insiders themselves realize what is happening at the far end of their distribution channels.

BrandLoyalties' clients receive daily metrics for over 3,800 global equities, which help to gauge the popularity of each company's brand names. In addition, BrandLoyalties uses proprietary normalization techniques to account for factors such as the growth in web traffic, bandwidth, and the proliferation of brand names in order to enable apples-to-apples comparisons. Ultimately, clients can use our YOY charts to help to identify when a company may be at risk or in position to pleasantly surprise on revenues.


How do you guys do it?

Collecting "Big Data" from cyberspace has become relatively easy, but capturing a robust and compliance friendly signal from all that noise has remained elusive. To derive our metrics we scan the web on a nightly basis, deconstruct or disassemble publicly available internet material into its component words, then accumulate, sort and count the frequency of the occurrences (or "citations") of each of the words. From this "lexicon" (which is simply a vocabulary or dictionary of the words we found on the internet) we extract those words that represent the over 45,000 brand names owned by the equities that we track. We are especially sensitive to compliance issues, and this lexicographic approach allows us to be fully compliant with the "fair use" provisions of the United States Copyright Act of 1976, 17 U.S.C. § 107. Additionally, our lexicographic process ensures that the data is totally anonymous and in full compliance with GDPR, CCPA or similar privacy regulations.

Once we have constructed a daily lexicon, we can measure the relative tallies of a company's products or services within the entire data set -- comparing the relative share of brand name citations with similar year-ago data and calculating the year-over-year growth (or contraction) in the share during each of the trailing 90 days. Those data points are then used to create regression slopes that are used to rank each equity relative to all others in the "universe" of equities that we track. The final result is a ranking based on percentiles, with lower percentiles (e.g., the 1st percentile) representing companies with strongly increasing popularity among on-line consumers.

Those relative rankings are expressed as "top X%" percentiles -- with lower numbers representing the better loyalties (e.g., the best 10% = 10th percentile and lower) and higher numbers the poorer loyalties (e.g. the worst 20% = 80th percentile and higher). Thus at any time the best half of the "full universe" of our tracked equities would be equities with percentile numbers of 50% or lower.


How long have you been doing this? How far back does your data go?

The quantitative BrandLoyalties rankings have been published to active clients daily since 2012 and are available proforma for dates as far back as 2008 for back-testing purposes. While the rankings are proprietary, they have been derived systematically and consistently since 2012 by using compliance-friendly "Big Data" scanning and mapping technologies which gauge online consumer interest relative to prior periods and to their peers.


What are the caveats to using your data?

Because of the on-line consumer data behind our BrandLoyalties metrics:

-- The metrics only work for companies with a strong on-line visibility -- something that we call "web luminosity."

-- They work best for companies whose principle operations are in the consumer sector -- a factor that we refer to as "signal relevance."

-- The accuracy of the "brand loyalties" signal generated is a function of both the "web luminosity" and "signal relevance" of any equity.

(For example, some equities have no presence in the consumer sector of the economy (e.g., LMT); others sell a lot of goods to consumer, but no transactions or significant portions of that commerce can be found on the web (e.g., XOM); still other equities have highly luminous sub-units operating on the web (e.g., BRK-A), but whether those units are good proxies for the health of the entire corporation requires much further analysis; in yet other cases the distribution channel is vertically segregated (e.g., KO/COKE or HOT/HST) and the importance of brand loyalties has to be understood in the context of where that brand loyalty most favorably impacts the operating units along the distribution channel; in some cases no amount of brand loyalty can overcome other structural, financial or management issues; and lastly some equities can become momentarily brilliant on the web for all of the wrong reasons.)

-- A positive Brand Loyalty ranking can mask suicidal pricing, a temporary novelty factor, product quality or service deterioration that haven't yet harmed consumers, or other fundamentally unsound business practices.

-- Real-world portfolios should only use BrandLoyalties metrics in combination with other research and common sense.

-- To provide a robust signal our metrics require a relatively high level of "web luminosity" -- making even our "full universe" of tracked equities highly selective. It typically includes only companies with "forward looking" distribution models and products or services that are in demand. Because of this, our "full universe" significantly out-performs the broader market indices -- as can be seen from the performance table on our "Home Page."

-- The lag-time between changes in BrandLoyalties and the consequential changes in earnings will depend on equity-specific details -- e.g., fiscal calendars and the length, complexity and inertia of the corporation's distribution channels.

-- We presently track over 4,500 globally traded stocks, and that number is increasing over time.

-- The data is very current -- daily data updated and posted on a daily basis.

-- And unlike other retail surveying technologies, the strength and accuracy of our signals will only improve as the web's share of total commerce inexorably increases.

Simply put: the "signal relevance" of any web-based data will change over time as consumers react to brands in varying ways -- and for varying reasons.


If the "Signal Relevance" of the BrandLoyalties data changes over time, how can we quantify and monitor it?

For each equity that we track we are able to calculate the correlation between consumer BrandLoyalties for the products of that corporation and subsequent revenues for the corporation -- and ultimately for subsequent movements of the equity's price. If corporate revenues (and therefore generally equity prices) increases with increasing brand loyalty, we will observe a positive mathematical correlation between corporate revenues, equity prices and our BrandLoyalties rankings.

On the other hand, if prices do not correlate at all to our BrandLoyalties rankings, we will observe a low or negative mathematical correlation between equity prices and our rankings. Thus our correlation data (for the trailing year and updated each day along with the rankings themselves) serves as a quantitative measure of the "Signal Relevance" for each of the equities that we track.

In addition to the raw correlation data, for our more sophisticated clients we also measure the time lag between changing BrandLoyalties and the consequential equity price movements. We do this by finding the time offset between the two series of data that demonstrate the best correlation. That data is also provided in our downloadable data sets.


What are the criteria for the rankings?

We measure the relative presence of a company's products or services within the lexicon of brand names that we have collected from "Big Data" each night -- comparing that relative level with similar year-ago data and calculate the changes in the relative demand over the past 90 days. Those data points are then used to rank each equity relative to all others in the "universe" of equities that we track. The final result is a ranking based on percentiles, with lower percentiles representing companies increasingly popular with on-line consumers.


Is your data measuring sentiment?

In the long term yes, but in the shorter term no. We only measure the frequency of brand name citations; we do not try to determine if any given reference to a brand is positive or negative. We are totally agnostic about positive or negative vibes; we only provide daily measurements of consumer awareness and interest in the brands that we cover.

The reason is simple: we do not trust sentiment parsing algorithms, especially when they are parsing social media -- which is replete with double and triple negatives, snarky comments and inconsistent emoticon usage. Additionally, the sampled demographics are self-selecting and therefore highly biased. For this reason we do not attempt to track moment-to-moment likes or dislikes.

However, over longer time frames (e.g., fiscal quarters) we are directly measuring consumer awareness and interest in the brands owned by a corporation. Changes in these kinds of measurements often anticipate changing revenue.

We also capture short term spikes in citation rates. We do not attempt to characterize those spikes as positive or negative for a corporation, we simply alert our clients that some kind of event is in progress. We expect our clients to have the resources to separate the excitement over a new product launch from a new public relations disaster.


From what web sites are your citations collected?

Our web crawling bots do not visit a persistent set of web pages. The bots simply start at randomly selected and link-rich publicly available pages and then follow the chain of embedded publicly available links wherever they may go. We do not even record the actual web addresses that we visit, only keeping cryptographic hashes that can tell us whether we have been there before and whether the pages have changed. We randomly select starting pages in order to sample as widely and thoroughly as possible.

We do not visit any page that requires access credentials of any kind, ensuring that all pages visited are public facing and therefore, by definition, absolutely void of material non-public information. Since most social media platforms require membership accounts and credentials, we do not directly sample from those platforms. However, large amounts of social media content is re-posted to publicly available web pages, which our bots are then likely to visit.

On the few occasions when we have actively monitored where our bots were crawling, we found that the collected brand citations were coming predominantly from crowd-sourced reviews. Only about 20% of the citations originated in social media.


When is your data collected? And how soon is it provided to your clients?

Our bots crawl the web during a fixed four hour time window each and every North American evening. This provides good bandwidth and latency in North America, excellent latency in Europe, but somewhat challenging latency in South East Asia. The fixed time window allows for readily standard-candle normalized samples, even under real-world material bandwidth and latency fluctuations.

The data collected each North American evening is processed into our metrics and delivered to clients before US markets open the following morning (e.i., corporate brand name citation metrics captured during day N are provided to subscribers on the morning of day N+1).

Our legacy data metrics are available for all days when the US markets have been open. Our newer AI/ML enhanced metrics are available for all calendar days.


Do you have data available for back-testing?

Yes, historic values of our BrandLoyalties metrics can be downloaded in either "proforma" or "point-in-time" data sets for back-testing analysis. The history files are provided in compressed CSV data formats.

Note that generally BrandLoyalties clients use our cleaner "Proforma" back data for initial model building or neural network training. They then confirm that model or network using the "out-of-sample" data provided by our "As Originally Published" (AKA: PIT "Point-in-Time") back data to test for the possibility of survivor bias in their results.

The possibility of survivor bias is more of an academic issue than a real-world portfolio issue when using data sets based on consumer brand loyalty. Because of our data collection methodology, our core equity coverage consists of highly liquid consumer companies with an actively engaged customer base. Such companies are rarely delisted because of failed business operations, and instead are usually "non-survivors" only because of M&A or other corporate actions -- which, in fact, generally result in portfolio gains. For this reason survivor bias within our back data is only likely to be found in portfolio models that focus on less liquid and smaller capitalization equities.

The "As Originally Published" data set is literally a concatenation of the daily files provided to our first client, and the feed for that client has remained unchanged over the years. Because of that it does not include all of the data fields in our current feeds or our "Proforma" data, and it does not correct for corporate actions, etc. Additionally, the PIT data does not reflect our current research on brand name mixes and our advanced standard-candle methodologies that better normalize for the inevitable bandwidth or latency impacted daily citation sample sizes.

On the other hand, our "Proforma" historic data file is intended as a real-time decision making aid for clients seeking to actively understand the historical performance of corporations during historic time frames with similar (or soon anticipated) economic circumstances. For that reason, those “historic” files are presented proforma, with the rankings of equities involved in merger/acquisitions or spin-offs presented as if the current equities had been historically represented by their current brand name mix.

These “proforma history” files are also proforma by virtue of utilizing the current brand name mix for each of the current equities in our universe – capturing newly covered equities and new product citation rates retrospectively and excluding delisted equities and discontinued products or brand names. Thus from time to time the mix of data sources, equities and brand names covered will vary for any given proforma date range -- causing the relative rankings and citation rates to change from an earlier proforma file to the latest proforma file for the same equity and time period. Note that ad hoc client requests for historic brand name or pre-IPO research are also prepared using the proforma approach. These files generally carry file names that are some variation of “BrandLoyalties_Full_History.”

Another difference between our “proforma” and our “as published” history files is that our early “as published” equity universe was materially smaller (~500 US equities) than our current universe (~2500 US equities, and an additional ~2500 Asian and European equities). That initial set of 500 equities was selected from among the largest capitalization corporations publicly traded in the US. For that reason the “proforma” data better represents the equity choices available currently, and it is arguably a more appropriate back-testing set than the much smaller “as published” data set.

Data for back-testing by prospective subscribers can be provided upon request after execution of a mutual NDA.


How often do you revise the list of brands that you are tracking?

We have a team that researches, reviews, updates and/or expands the brand names in our coverage universe on a daily basis. Obviously this means that the brands that we track for any given ticker X may change from day to day. For this reason we recommend that any production trading algorithms utilize our most recent daily "proforma" data histories, which recasts all of our metrics historically using the latest brand mixes for each of our covered equities.


What equities are in the BrandLoyalties "universe"?

The number of equities being tracked is constantly changing, with new equities being added all the time. Our "Home" page at www.BrandLoyalties.com contains the most current count of tracked equities, and displays lists of those equities sorted by ticker symbol and each equity's industry. Separate lists are provided for our European and Asian coverage, and a link is provided for downloading the entire list of equities that we track in a simple spreadsheet. Some of those equities have very noisy citation rates and are excluded from our daily metrics; the list of equities with good enough signals to be included in last night's metrics can be downloaded in this spreadsheet. A complete history of our equity coverage can be downloaded here.


In what data formats are the metrics available?

The BrandLoyalties metrics are available to subscribers in CSV file formats that can be downloaded or transferred via FTP (either push or pull).


Do you have any suggestions on the best way to get started when using Artificial Intelligence and Machine Learning with your data?

"Newbies" should probably start with a Python based Neural Network platform (e.g. scikit-learn v0.20.0) before moving to Google's Tensor Flow for production purposes.

Model each of our metrics in single factor models and determine relative importance weightings (using Garson's algorithm), before attempting to build final multi-factor models.

Train the model (e.g. sklearn.neural_network.MLPClassifier) on rolling two year periods with the goal of predicting the following 91 day ticker-by-ticker excess return quintiles.

Use the ticker-by-ticker predicted excess returns quintiles to overweight / underweight equities in model portfolio.


How effective are the BrandLoyalties metrics?

The BrandLoyalties "Home" page contains a table that compares the performance of a number of custom "Smart Beta" indexes that use BrandLoyalties metrics as the sole weighting criteria. The indexes shown include various market capitalization tranches, broad consumer industries and narrower consumer sectors. Links are provided to download historical daily index values and "Fact Sheets" for each of the indexes. Additionally, we provide a link to a Data Usage Example PDF that describes in detail the methodology used to construct the index portfolios.


What other information do you provide?

The top portion of the BrandLoyalties "Home" page contains links to several additional resources:

-- Our latest "White Paper" which contains a detailed description of our methodologies and processes.

-- A YouTube video version of our current Webinar presentation.

-- A link to download a spreadsheet containing our current full "universe" of equity coverage.

Additionally, the peer-reviewed article in the Summer 2015 Journal of Index Investing "Big Data" Meets "Smart Beta" is a primer for investment professionals about how to glean relevant insights from brand-name metrics, and the June 2019 Journal of Investing article "Using Alternative Research Data in Real-World Portfolios" provides detailed analysis of how high performance portfolios can be constituted using our data.


What is BrandLoyalties, Inc. and what is its history?

A sample Due Diligence Questionnaire containing a summary of that information can be downloaded here.


How are the BrandLoyalties subscription plans structured and priced?

Please contact Tony Seker for more information on available subscription plans and pricing. You may contact him as shown below:



(Click here if image is not visible)

---

Home

Basic Concepts White Paper

3 Minute Overview Video

BrandLoyalties Full Presentation

"Big Data Meets Deep Learning" Video

BrandLoyalties FAQs

---

Copyright �2024 BrandLoyalties, Inc.